[R] [R-sig-ME] Duplicating meta-regression results from PROC MIXED with lmer

[Brant Inman]

Rolf,

I actually don't believe that this is a SAS vs R issue since I have 3  
sources that report the same results.  I know that STATA, SAS and the  
mima function from R can all be used to give the correct results.  The  
question is related more to how I can get similar results with lmer.   
Currently, the code I provided generates VERY different results,  
especially related to between study variance estimation (tau) and  
regression coefficient standard errors. Basically, I think I got it  
all wrong in my coding of the models with covariates (f3 and f4), but  
I can't understand why.

Brant


On May 5, 2009, at 11:43 PM, Rolf Turner wrote:

>
> On 6/05/2009, at 3:00 PM, Brant Inman wrote:
>
>> R-experts:
>>
>> In 2002, Hans Van Houwelingen et al. published a tutorial on how to  
>> do
>> meta-regression in Statistics in Medicine.  They used the classic BCG
>> dataset of Colditz to demonstrate correct methodology and computed  
>> the
>> results using PROC MIXED in SAS. In trying to duplicate the results
>> presented in this paper, I have discovered that I can reproduce
>> certain items with lmer but not others.  I was hoping that someone
>> might point out how I could correctly program R code to arrive at the
>> correct solution.
>
> 	<snip>
>
> There appears to be a tacit assertion here that the results from PROC
> MIXED in The-Package-That-Must-Not-Be-Named are the correct results.
>
> This assertion is very likely to bring the wrath of Doug Bates down
> upon your head.  An outcome to be devoutly avoided! :-)
>
> 	cheers,
>
> 		Rolf Turner
>
> ######################################################################
> Attention: This e-mail message is privileged and confidential. If  
> you are not the intended recipient please delete the message and  
> notify the sender. Any views or opinions presented are solely those  
> of the author.
>
> This e-mail has been scanned and cleared by MailMarshal www.marshalsoftware.com
> ######################################################################